Heuristic construction of exact experimental designs under multiple resource constraints
The aim of this paper is twofold. First, we introduce "resource constraints" as a general concept that covers many practical restrictions on experimental design. Second, for computing efficient exact designs of experiments under any combination of resource constraints, we propose a tabu search heuristic that uses some ideas of the Detmax procedure. To illustrate the scope and performance of our heuristic, we computed D-efficient designs for 1) a block model with limits on the numbers of blocks and on the availability of experimental material; 2) a quadratic regression model with simultaneous marginal and cost constraints; 3) a non-linear regression model with simultaneous direct and cost constraints. As we show, the proposed heuristic generates comparable or better results than algorithms specialized for computing optimal designs under less general constraints.